Semantic Role Labeling (SRL), or shallow semantic parsing, is the task of identifying the predicate (action) in a sentence and classifying its arguments into semantic roles (Agent, Patient, Instrument, Locative) — answering "Who did what to whom, where, when, and how?".

PropBank Roles

• Arg0 (Proto-Agent): The doer/cause (Subject).
• Arg1 (Proto-Patient): The thing affected (Object).
• Arg2-Arg5: Verb-specific roles (Beneficiary, Start Point, End Point).
• Adjuncts: Time (AM-TMP), Location (AM-LOC), Manner (AM-MNR).

Example

• "John (Arg0) broke the window (Arg1) with a rock (Arg3) yesterday (AM-TMP)."

Why It Matters

• Abstraction: "The window was broken by John" and "John broke the window" have different syntax but IDENTICAL Semantic Roles.
• QA: Directly maps natural language questions to structured database queries.
• Event Extraction: The core component of event extraction systems.

Semantic Role Labeling is normalizing meaning — disregarding passive/active voice or word order to extract the core event structure.